1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/errno.h> 21 #include <linux/types.h> 22 #include <linux/socket.h> 23 #include <linux/sockios.h> 24 #include <linux/net.h> 25 #include <linux/in6.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_arp.h> 28 #include <linux/ipv6.h> 29 #include <linux/icmpv6.h> 30 #include <linux/init.h> 31 #include <linux/module.h> 32 #include <linux/skbuff.h> 33 #include <linux/slab.h> 34 #include <linux/uaccess.h> 35 #include <linux/indirect_call_wrapper.h> 36 37 #include <net/addrconf.h> 38 #include <net/ndisc.h> 39 #include <net/protocol.h> 40 #include <net/transp_v6.h> 41 #include <net/ip6_route.h> 42 #include <net/raw.h> 43 #include <net/tcp_states.h> 44 #include <net/ip6_checksum.h> 45 #include <net/ip6_tunnel.h> 46 #include <net/xfrm.h> 47 #include <net/inet_hashtables.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/busy_poll.h> 50 #include <net/sock_reuseport.h> 51 52 #include <linux/proc_fs.h> 53 #include <linux/seq_file.h> 54 #include <trace/events/skb.h> 55 #include "udp_impl.h" 56 57 static u32 udp6_ehashfn(const struct net *net, 58 const struct in6_addr *laddr, 59 const u16 lport, 60 const struct in6_addr *faddr, 61 const __be16 fport) 62 { 63 static u32 udp6_ehash_secret __read_mostly; 64 static u32 udp_ipv6_hash_secret __read_mostly; 65 66 u32 lhash, fhash; 67 68 net_get_random_once(&udp6_ehash_secret, 69 sizeof(udp6_ehash_secret)); 70 net_get_random_once(&udp_ipv6_hash_secret, 71 sizeof(udp_ipv6_hash_secret)); 72 73 lhash = (__force u32)laddr->s6_addr32[3]; 74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 75 76 return __inet6_ehashfn(lhash, lport, fhash, fport, 77 udp_ipv6_hash_secret + net_hash_mix(net)); 78 } 79 80 int udp_v6_get_port(struct sock *sk, unsigned short snum) 81 { 82 unsigned int hash2_nulladdr = 83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 84 unsigned int hash2_partial = 85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 86 87 /* precompute partial secondary hash */ 88 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 89 return udp_lib_get_port(sk, snum, hash2_nulladdr); 90 } 91 92 void udp_v6_rehash(struct sock *sk) 93 { 94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 95 &sk->sk_v6_rcv_saddr, 96 inet_sk(sk)->inet_num); 97 98 udp_lib_rehash(sk, new_hash); 99 } 100 101 static int compute_score(struct sock *sk, struct net *net, 102 const struct in6_addr *saddr, __be16 sport, 103 const struct in6_addr *daddr, unsigned short hnum, 104 int dif, int sdif) 105 { 106 int score; 107 struct inet_sock *inet; 108 bool dev_match; 109 110 if (!net_eq(sock_net(sk), net) || 111 udp_sk(sk)->udp_port_hash != hnum || 112 sk->sk_family != PF_INET6) 113 return -1; 114 115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 116 return -1; 117 118 score = 0; 119 inet = inet_sk(sk); 120 121 if (inet->inet_dport) { 122 if (inet->inet_dport != sport) 123 return -1; 124 score++; 125 } 126 127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 129 return -1; 130 score++; 131 } 132 133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 134 if (!dev_match) 135 return -1; 136 score++; 137 138 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 139 score++; 140 141 return score; 142 } 143 144 /* called with rcu_read_lock() */ 145 static struct sock *udp6_lib_lookup2(struct net *net, 146 const struct in6_addr *saddr, __be16 sport, 147 const struct in6_addr *daddr, unsigned int hnum, 148 int dif, int sdif, struct udp_hslot *hslot2, 149 struct sk_buff *skb) 150 { 151 struct sock *sk, *result; 152 int score, badness; 153 u32 hash = 0; 154 155 result = NULL; 156 badness = -1; 157 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 158 score = compute_score(sk, net, saddr, sport, 159 daddr, hnum, dif, sdif); 160 if (score > badness) { 161 if (sk->sk_reuseport) { 162 hash = udp6_ehashfn(net, daddr, hnum, 163 saddr, sport); 164 165 result = reuseport_select_sock(sk, hash, skb, 166 sizeof(struct udphdr)); 167 if (result) 168 return result; 169 } 170 result = sk; 171 badness = score; 172 } 173 } 174 return result; 175 } 176 177 /* rcu_read_lock() must be held */ 178 struct sock *__udp6_lib_lookup(struct net *net, 179 const struct in6_addr *saddr, __be16 sport, 180 const struct in6_addr *daddr, __be16 dport, 181 int dif, int sdif, struct udp_table *udptable, 182 struct sk_buff *skb) 183 { 184 unsigned short hnum = ntohs(dport); 185 unsigned int hash2, slot2; 186 struct udp_hslot *hslot2; 187 struct sock *result; 188 189 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 190 slot2 = hash2 & udptable->mask; 191 hslot2 = &udptable->hash2[slot2]; 192 193 result = udp6_lib_lookup2(net, saddr, sport, 194 daddr, hnum, dif, sdif, 195 hslot2, skb); 196 if (!result) { 197 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 198 slot2 = hash2 & udptable->mask; 199 200 hslot2 = &udptable->hash2[slot2]; 201 202 result = udp6_lib_lookup2(net, saddr, sport, 203 &in6addr_any, hnum, dif, sdif, 204 hslot2, skb); 205 } 206 if (IS_ERR(result)) 207 return NULL; 208 return result; 209 } 210 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 211 212 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 213 __be16 sport, __be16 dport, 214 struct udp_table *udptable) 215 { 216 const struct ipv6hdr *iph = ipv6_hdr(skb); 217 218 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 219 &iph->daddr, dport, inet6_iif(skb), 220 inet6_sdif(skb), udptable, skb); 221 } 222 223 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 224 __be16 sport, __be16 dport) 225 { 226 const struct ipv6hdr *iph = ipv6_hdr(skb); 227 228 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 229 &iph->daddr, dport, inet6_iif(skb), 230 inet6_sdif(skb), &udp_table, skb); 231 } 232 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 233 234 /* Must be called under rcu_read_lock(). 235 * Does increment socket refcount. 236 */ 237 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 238 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 239 const struct in6_addr *daddr, __be16 dport, int dif) 240 { 241 struct sock *sk; 242 243 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 244 dif, 0, &udp_table, NULL); 245 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 246 sk = NULL; 247 return sk; 248 } 249 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 250 #endif 251 252 /* do not use the scratch area len for jumbogram: their length execeeds the 253 * scratch area space; note that the IP6CB flags is still in the first 254 * cacheline, so checking for jumbograms is cheap 255 */ 256 static int udp6_skb_len(struct sk_buff *skb) 257 { 258 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 259 } 260 261 /* 262 * This should be easy, if there is something there we 263 * return it, otherwise we block. 264 */ 265 266 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 267 int noblock, int flags, int *addr_len) 268 { 269 struct ipv6_pinfo *np = inet6_sk(sk); 270 struct inet_sock *inet = inet_sk(sk); 271 struct sk_buff *skb; 272 unsigned int ulen, copied; 273 int off, err, peeking = flags & MSG_PEEK; 274 int is_udplite = IS_UDPLITE(sk); 275 struct udp_mib __percpu *mib; 276 bool checksum_valid = false; 277 int is_udp4; 278 279 if (flags & MSG_ERRQUEUE) 280 return ipv6_recv_error(sk, msg, len, addr_len); 281 282 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 283 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 284 285 try_again: 286 off = sk_peek_offset(sk, flags); 287 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 288 if (!skb) 289 return err; 290 291 ulen = udp6_skb_len(skb); 292 copied = len; 293 if (copied > ulen - off) 294 copied = ulen - off; 295 else if (copied < ulen) 296 msg->msg_flags |= MSG_TRUNC; 297 298 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 299 mib = __UDPX_MIB(sk, is_udp4); 300 301 /* 302 * If checksum is needed at all, try to do it while copying the 303 * data. If the data is truncated, or if we only want a partial 304 * coverage checksum (UDP-Lite), do it before the copy. 305 */ 306 307 if (copied < ulen || peeking || 308 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 309 checksum_valid = udp_skb_csum_unnecessary(skb) || 310 !__udp_lib_checksum_complete(skb); 311 if (!checksum_valid) 312 goto csum_copy_err; 313 } 314 315 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 316 if (udp_skb_is_linear(skb)) 317 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 318 else 319 err = skb_copy_datagram_msg(skb, off, msg, copied); 320 } else { 321 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 322 if (err == -EINVAL) 323 goto csum_copy_err; 324 } 325 if (unlikely(err)) { 326 if (!peeking) { 327 atomic_inc(&sk->sk_drops); 328 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 329 } 330 kfree_skb(skb); 331 return err; 332 } 333 if (!peeking) 334 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 335 336 sock_recv_ts_and_drops(msg, sk, skb); 337 338 /* Copy the address. */ 339 if (msg->msg_name) { 340 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 341 sin6->sin6_family = AF_INET6; 342 sin6->sin6_port = udp_hdr(skb)->source; 343 sin6->sin6_flowinfo = 0; 344 345 if (is_udp4) { 346 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 347 &sin6->sin6_addr); 348 sin6->sin6_scope_id = 0; 349 } else { 350 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 351 sin6->sin6_scope_id = 352 ipv6_iface_scope_id(&sin6->sin6_addr, 353 inet6_iif(skb)); 354 } 355 *addr_len = sizeof(*sin6); 356 } 357 358 if (udp_sk(sk)->gro_enabled) 359 udp_cmsg_recv(msg, sk, skb); 360 361 if (np->rxopt.all) 362 ip6_datagram_recv_common_ctl(sk, msg, skb); 363 364 if (is_udp4) { 365 if (inet->cmsg_flags) 366 ip_cmsg_recv_offset(msg, sk, skb, 367 sizeof(struct udphdr), off); 368 } else { 369 if (np->rxopt.all) 370 ip6_datagram_recv_specific_ctl(sk, msg, skb); 371 } 372 373 err = copied; 374 if (flags & MSG_TRUNC) 375 err = ulen; 376 377 skb_consume_udp(sk, skb, peeking ? -err : err); 378 return err; 379 380 csum_copy_err: 381 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 382 udp_skb_destructor)) { 383 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 384 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 385 } 386 kfree_skb(skb); 387 388 /* starting over for a new packet, but check if we need to yield */ 389 cond_resched(); 390 msg->msg_flags &= ~MSG_TRUNC; 391 goto try_again; 392 } 393 394 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 395 void udpv6_encap_enable(void) 396 { 397 static_branch_inc(&udpv6_encap_needed_key); 398 } 399 EXPORT_SYMBOL(udpv6_encap_enable); 400 401 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 402 * through error handlers in encapsulations looking for a match. 403 */ 404 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 405 struct inet6_skb_parm *opt, 406 u8 type, u8 code, int offset, __be32 info) 407 { 408 int i; 409 410 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 411 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 412 u8 type, u8 code, int offset, __be32 info); 413 const struct ip6_tnl_encap_ops *encap; 414 415 encap = rcu_dereference(ip6tun_encaps[i]); 416 if (!encap) 417 continue; 418 handler = encap->err_handler; 419 if (handler && !handler(skb, opt, type, code, offset, info)) 420 return 0; 421 } 422 423 return -ENOENT; 424 } 425 426 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 427 * reversing source and destination port: this will match tunnels that force the 428 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 429 * lwtunnels might actually break this assumption by being configured with 430 * different destination ports on endpoints, in this case we won't be able to 431 * trace ICMP messages back to them. 432 * 433 * If this doesn't match any socket, probe tunnels with arbitrary destination 434 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 435 * we've sent packets to won't necessarily match the local destination port. 436 * 437 * Then ask the tunnel implementation to match the error against a valid 438 * association. 439 * 440 * Return an error if we can't find a match, the socket if we need further 441 * processing, zero otherwise. 442 */ 443 static struct sock *__udp6_lib_err_encap(struct net *net, 444 const struct ipv6hdr *hdr, int offset, 445 struct udphdr *uh, 446 struct udp_table *udptable, 447 struct sk_buff *skb, 448 struct inet6_skb_parm *opt, 449 u8 type, u8 code, __be32 info) 450 { 451 int network_offset, transport_offset; 452 struct sock *sk; 453 454 network_offset = skb_network_offset(skb); 455 transport_offset = skb_transport_offset(skb); 456 457 /* Network header needs to point to the outer IPv6 header inside ICMP */ 458 skb_reset_network_header(skb); 459 460 /* Transport header needs to point to the UDP header */ 461 skb_set_transport_header(skb, offset); 462 463 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 464 &hdr->saddr, uh->dest, 465 inet6_iif(skb), 0, udptable, skb); 466 if (sk) { 467 int (*lookup)(struct sock *sk, struct sk_buff *skb); 468 struct udp_sock *up = udp_sk(sk); 469 470 lookup = READ_ONCE(up->encap_err_lookup); 471 if (!lookup || lookup(sk, skb)) 472 sk = NULL; 473 } 474 475 if (!sk) { 476 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 477 offset, info)); 478 } 479 480 skb_set_transport_header(skb, transport_offset); 481 skb_set_network_header(skb, network_offset); 482 483 return sk; 484 } 485 486 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 487 u8 type, u8 code, int offset, __be32 info, 488 struct udp_table *udptable) 489 { 490 struct ipv6_pinfo *np; 491 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 492 const struct in6_addr *saddr = &hdr->saddr; 493 const struct in6_addr *daddr = &hdr->daddr; 494 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 495 bool tunnel = false; 496 struct sock *sk; 497 int harderr; 498 int err; 499 struct net *net = dev_net(skb->dev); 500 501 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 502 inet6_iif(skb), inet6_sdif(skb), udptable, skb); 503 if (!sk) { 504 /* No socket for error: try tunnels before discarding */ 505 sk = ERR_PTR(-ENOENT); 506 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 507 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 508 udptable, skb, 509 opt, type, code, info); 510 if (!sk) 511 return 0; 512 } 513 514 if (IS_ERR(sk)) { 515 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 516 ICMP6_MIB_INERRORS); 517 return PTR_ERR(sk); 518 } 519 520 tunnel = true; 521 } 522 523 harderr = icmpv6_err_convert(type, code, &err); 524 np = inet6_sk(sk); 525 526 if (type == ICMPV6_PKT_TOOBIG) { 527 if (!ip6_sk_accept_pmtu(sk)) 528 goto out; 529 ip6_sk_update_pmtu(skb, sk, info); 530 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 531 harderr = 1; 532 } 533 if (type == NDISC_REDIRECT) { 534 if (tunnel) { 535 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 536 sk->sk_mark, sk->sk_uid); 537 } else { 538 ip6_sk_redirect(skb, sk); 539 } 540 goto out; 541 } 542 543 /* Tunnels don't have an application socket: don't pass errors back */ 544 if (tunnel) 545 goto out; 546 547 if (!np->recverr) { 548 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 549 goto out; 550 } else { 551 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 552 } 553 554 sk->sk_err = err; 555 sk->sk_error_report(sk); 556 out: 557 return 0; 558 } 559 560 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 561 { 562 int rc; 563 564 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 565 sock_rps_save_rxhash(sk, skb); 566 sk_mark_napi_id(sk, skb); 567 sk_incoming_cpu_update(sk); 568 } else { 569 sk_mark_napi_id_once(sk, skb); 570 } 571 572 rc = __udp_enqueue_schedule_skb(sk, skb); 573 if (rc < 0) { 574 int is_udplite = IS_UDPLITE(sk); 575 576 /* Note that an ENOMEM error is charged twice */ 577 if (rc == -ENOMEM) 578 UDP6_INC_STATS(sock_net(sk), 579 UDP_MIB_RCVBUFERRORS, is_udplite); 580 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 581 kfree_skb(skb); 582 return -1; 583 } 584 585 return 0; 586 } 587 588 static __inline__ int udpv6_err(struct sk_buff *skb, 589 struct inet6_skb_parm *opt, u8 type, 590 u8 code, int offset, __be32 info) 591 { 592 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 593 } 594 595 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 596 { 597 struct udp_sock *up = udp_sk(sk); 598 int is_udplite = IS_UDPLITE(sk); 599 600 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 601 goto drop; 602 603 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 604 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 605 606 /* 607 * This is an encapsulation socket so pass the skb to 608 * the socket's udp_encap_rcv() hook. Otherwise, just 609 * fall through and pass this up the UDP socket. 610 * up->encap_rcv() returns the following value: 611 * =0 if skb was successfully passed to the encap 612 * handler or was discarded by it. 613 * >0 if skb should be passed on to UDP. 614 * <0 if skb should be resubmitted as proto -N 615 */ 616 617 /* if we're overly short, let UDP handle it */ 618 encap_rcv = READ_ONCE(up->encap_rcv); 619 if (encap_rcv) { 620 int ret; 621 622 /* Verify checksum before giving to encap */ 623 if (udp_lib_checksum_complete(skb)) 624 goto csum_error; 625 626 ret = encap_rcv(sk, skb); 627 if (ret <= 0) { 628 __UDP_INC_STATS(sock_net(sk), 629 UDP_MIB_INDATAGRAMS, 630 is_udplite); 631 return -ret; 632 } 633 } 634 635 /* FALLTHROUGH -- it's a UDP Packet */ 636 } 637 638 /* 639 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 640 */ 641 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 642 643 if (up->pcrlen == 0) { /* full coverage was set */ 644 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 645 UDP_SKB_CB(skb)->cscov, skb->len); 646 goto drop; 647 } 648 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 649 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 650 UDP_SKB_CB(skb)->cscov, up->pcrlen); 651 goto drop; 652 } 653 } 654 655 prefetch(&sk->sk_rmem_alloc); 656 if (rcu_access_pointer(sk->sk_filter) && 657 udp_lib_checksum_complete(skb)) 658 goto csum_error; 659 660 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 661 goto drop; 662 663 udp_csum_pull_header(skb); 664 665 skb_dst_drop(skb); 666 667 return __udpv6_queue_rcv_skb(sk, skb); 668 669 csum_error: 670 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 671 drop: 672 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 673 atomic_inc(&sk->sk_drops); 674 kfree_skb(skb); 675 return -1; 676 } 677 678 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 679 { 680 struct sk_buff *next, *segs; 681 int ret; 682 683 if (likely(!udp_unexpected_gso(sk, skb))) 684 return udpv6_queue_rcv_one_skb(sk, skb); 685 686 __skb_push(skb, -skb_mac_offset(skb)); 687 segs = udp_rcv_segment(sk, skb, false); 688 for (skb = segs; skb; skb = next) { 689 next = skb->next; 690 __skb_pull(skb, skb_transport_offset(skb)); 691 692 ret = udpv6_queue_rcv_one_skb(sk, skb); 693 if (ret > 0) 694 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 695 true); 696 } 697 return 0; 698 } 699 700 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 701 __be16 loc_port, const struct in6_addr *loc_addr, 702 __be16 rmt_port, const struct in6_addr *rmt_addr, 703 int dif, int sdif, unsigned short hnum) 704 { 705 struct inet_sock *inet = inet_sk(sk); 706 707 if (!net_eq(sock_net(sk), net)) 708 return false; 709 710 if (udp_sk(sk)->udp_port_hash != hnum || 711 sk->sk_family != PF_INET6 || 712 (inet->inet_dport && inet->inet_dport != rmt_port) || 713 (!ipv6_addr_any(&sk->sk_v6_daddr) && 714 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 715 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 716 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 717 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 718 return false; 719 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 720 return false; 721 return true; 722 } 723 724 static void udp6_csum_zero_error(struct sk_buff *skb) 725 { 726 /* RFC 2460 section 8.1 says that we SHOULD log 727 * this error. Well, it is reasonable. 728 */ 729 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 730 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 731 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 732 } 733 734 /* 735 * Note: called only from the BH handler context, 736 * so we don't need to lock the hashes. 737 */ 738 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 739 const struct in6_addr *saddr, const struct in6_addr *daddr, 740 struct udp_table *udptable, int proto) 741 { 742 struct sock *sk, *first = NULL; 743 const struct udphdr *uh = udp_hdr(skb); 744 unsigned short hnum = ntohs(uh->dest); 745 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 746 unsigned int offset = offsetof(typeof(*sk), sk_node); 747 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 748 int dif = inet6_iif(skb); 749 int sdif = inet6_sdif(skb); 750 struct hlist_node *node; 751 struct sk_buff *nskb; 752 753 if (use_hash2) { 754 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 755 udptable->mask; 756 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 757 start_lookup: 758 hslot = &udptable->hash2[hash2]; 759 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 760 } 761 762 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 763 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 764 uh->source, saddr, dif, sdif, 765 hnum)) 766 continue; 767 /* If zero checksum and no_check is not on for 768 * the socket then skip it. 769 */ 770 if (!uh->check && !udp_sk(sk)->no_check6_rx) 771 continue; 772 if (!first) { 773 first = sk; 774 continue; 775 } 776 nskb = skb_clone(skb, GFP_ATOMIC); 777 if (unlikely(!nskb)) { 778 atomic_inc(&sk->sk_drops); 779 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 780 IS_UDPLITE(sk)); 781 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 782 IS_UDPLITE(sk)); 783 continue; 784 } 785 786 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 787 consume_skb(nskb); 788 } 789 790 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 791 if (use_hash2 && hash2 != hash2_any) { 792 hash2 = hash2_any; 793 goto start_lookup; 794 } 795 796 if (first) { 797 if (udpv6_queue_rcv_skb(first, skb) > 0) 798 consume_skb(skb); 799 } else { 800 kfree_skb(skb); 801 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 802 proto == IPPROTO_UDPLITE); 803 } 804 return 0; 805 } 806 807 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 808 { 809 if (udp_sk_rx_dst_set(sk, dst)) { 810 const struct rt6_info *rt = (const struct rt6_info *)dst; 811 812 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 813 } 814 } 815 816 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 817 * return code conversion for ip layer consumption 818 */ 819 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 820 struct udphdr *uh) 821 { 822 int ret; 823 824 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 825 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 826 ip6_compute_pseudo); 827 828 ret = udpv6_queue_rcv_skb(sk, skb); 829 830 /* a return value > 0 means to resubmit the input */ 831 if (ret > 0) 832 return ret; 833 return 0; 834 } 835 836 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 837 int proto) 838 { 839 const struct in6_addr *saddr, *daddr; 840 struct net *net = dev_net(skb->dev); 841 struct udphdr *uh; 842 struct sock *sk; 843 u32 ulen = 0; 844 845 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 846 goto discard; 847 848 saddr = &ipv6_hdr(skb)->saddr; 849 daddr = &ipv6_hdr(skb)->daddr; 850 uh = udp_hdr(skb); 851 852 ulen = ntohs(uh->len); 853 if (ulen > skb->len) 854 goto short_packet; 855 856 if (proto == IPPROTO_UDP) { 857 /* UDP validates ulen. */ 858 859 /* Check for jumbo payload */ 860 if (ulen == 0) 861 ulen = skb->len; 862 863 if (ulen < sizeof(*uh)) 864 goto short_packet; 865 866 if (ulen < skb->len) { 867 if (pskb_trim_rcsum(skb, ulen)) 868 goto short_packet; 869 saddr = &ipv6_hdr(skb)->saddr; 870 daddr = &ipv6_hdr(skb)->daddr; 871 uh = udp_hdr(skb); 872 } 873 } 874 875 if (udp6_csum_init(skb, uh, proto)) 876 goto csum_error; 877 878 /* Check if the socket is already available, e.g. due to early demux */ 879 sk = skb_steal_sock(skb); 880 if (sk) { 881 struct dst_entry *dst = skb_dst(skb); 882 int ret; 883 884 if (unlikely(sk->sk_rx_dst != dst)) 885 udp6_sk_rx_dst_set(sk, dst); 886 887 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 888 sock_put(sk); 889 goto report_csum_error; 890 } 891 892 ret = udp6_unicast_rcv_skb(sk, skb, uh); 893 sock_put(sk); 894 return ret; 895 } 896 897 /* 898 * Multicast receive code 899 */ 900 if (ipv6_addr_is_multicast(daddr)) 901 return __udp6_lib_mcast_deliver(net, skb, 902 saddr, daddr, udptable, proto); 903 904 /* Unicast */ 905 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 906 if (sk) { 907 if (!uh->check && !udp_sk(sk)->no_check6_rx) 908 goto report_csum_error; 909 return udp6_unicast_rcv_skb(sk, skb, uh); 910 } 911 912 if (!uh->check) 913 goto report_csum_error; 914 915 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 916 goto discard; 917 918 if (udp_lib_checksum_complete(skb)) 919 goto csum_error; 920 921 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 922 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 923 924 kfree_skb(skb); 925 return 0; 926 927 short_packet: 928 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 929 proto == IPPROTO_UDPLITE ? "-Lite" : "", 930 saddr, ntohs(uh->source), 931 ulen, skb->len, 932 daddr, ntohs(uh->dest)); 933 goto discard; 934 935 report_csum_error: 936 udp6_csum_zero_error(skb); 937 csum_error: 938 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 939 discard: 940 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 941 kfree_skb(skb); 942 return 0; 943 } 944 945 946 static struct sock *__udp6_lib_demux_lookup(struct net *net, 947 __be16 loc_port, const struct in6_addr *loc_addr, 948 __be16 rmt_port, const struct in6_addr *rmt_addr, 949 int dif, int sdif) 950 { 951 unsigned short hnum = ntohs(loc_port); 952 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 953 unsigned int slot2 = hash2 & udp_table.mask; 954 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 955 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 956 struct sock *sk; 957 958 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 959 if (sk->sk_state == TCP_ESTABLISHED && 960 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 961 return sk; 962 /* Only check first socket in chain */ 963 break; 964 } 965 return NULL; 966 } 967 968 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 969 { 970 struct net *net = dev_net(skb->dev); 971 const struct udphdr *uh; 972 struct sock *sk; 973 struct dst_entry *dst; 974 int dif = skb->dev->ifindex; 975 int sdif = inet6_sdif(skb); 976 977 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 978 sizeof(struct udphdr))) 979 return; 980 981 uh = udp_hdr(skb); 982 983 if (skb->pkt_type == PACKET_HOST) 984 sk = __udp6_lib_demux_lookup(net, uh->dest, 985 &ipv6_hdr(skb)->daddr, 986 uh->source, &ipv6_hdr(skb)->saddr, 987 dif, sdif); 988 else 989 return; 990 991 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 992 return; 993 994 skb->sk = sk; 995 skb->destructor = sock_efree; 996 dst = READ_ONCE(sk->sk_rx_dst); 997 998 if (dst) 999 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1000 if (dst) { 1001 /* set noref for now. 1002 * any place which wants to hold dst has to call 1003 * dst_hold_safe() 1004 */ 1005 skb_dst_set_noref(skb, dst); 1006 } 1007 } 1008 1009 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1010 { 1011 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1012 } 1013 1014 /* 1015 * Throw away all pending data and cancel the corking. Socket is locked. 1016 */ 1017 static void udp_v6_flush_pending_frames(struct sock *sk) 1018 { 1019 struct udp_sock *up = udp_sk(sk); 1020 1021 if (up->pending == AF_INET) 1022 udp_flush_pending_frames(sk); 1023 else if (up->pending) { 1024 up->len = 0; 1025 up->pending = 0; 1026 ip6_flush_pending_frames(sk); 1027 } 1028 } 1029 1030 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1031 int addr_len) 1032 { 1033 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1034 return -EINVAL; 1035 /* The following checks are replicated from __ip6_datagram_connect() 1036 * and intended to prevent BPF program called below from accessing 1037 * bytes that are out of the bound specified by user in addr_len. 1038 */ 1039 if (uaddr->sa_family == AF_INET) { 1040 if (__ipv6_only_sock(sk)) 1041 return -EAFNOSUPPORT; 1042 return udp_pre_connect(sk, uaddr, addr_len); 1043 } 1044 1045 if (addr_len < SIN6_LEN_RFC2133) 1046 return -EINVAL; 1047 1048 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1049 } 1050 1051 /** 1052 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1053 * @sk: socket we are sending on 1054 * @skb: sk_buff containing the filled-in UDP header 1055 * (checksum field must be zeroed out) 1056 */ 1057 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1058 const struct in6_addr *saddr, 1059 const struct in6_addr *daddr, int len) 1060 { 1061 unsigned int offset; 1062 struct udphdr *uh = udp_hdr(skb); 1063 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1064 __wsum csum = 0; 1065 1066 if (!frags) { 1067 /* Only one fragment on the socket. */ 1068 skb->csum_start = skb_transport_header(skb) - skb->head; 1069 skb->csum_offset = offsetof(struct udphdr, check); 1070 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1071 } else { 1072 /* 1073 * HW-checksum won't work as there are two or more 1074 * fragments on the socket so that all csums of sk_buffs 1075 * should be together 1076 */ 1077 offset = skb_transport_offset(skb); 1078 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1079 csum = skb->csum; 1080 1081 skb->ip_summed = CHECKSUM_NONE; 1082 1083 do { 1084 csum = csum_add(csum, frags->csum); 1085 } while ((frags = frags->next)); 1086 1087 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1088 csum); 1089 if (uh->check == 0) 1090 uh->check = CSUM_MANGLED_0; 1091 } 1092 } 1093 1094 /* 1095 * Sending 1096 */ 1097 1098 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1099 struct inet_cork *cork) 1100 { 1101 struct sock *sk = skb->sk; 1102 struct udphdr *uh; 1103 int err = 0; 1104 int is_udplite = IS_UDPLITE(sk); 1105 __wsum csum = 0; 1106 int offset = skb_transport_offset(skb); 1107 int len = skb->len - offset; 1108 1109 /* 1110 * Create a UDP header 1111 */ 1112 uh = udp_hdr(skb); 1113 uh->source = fl6->fl6_sport; 1114 uh->dest = fl6->fl6_dport; 1115 uh->len = htons(len); 1116 uh->check = 0; 1117 1118 if (cork->gso_size) { 1119 const int hlen = skb_network_header_len(skb) + 1120 sizeof(struct udphdr); 1121 1122 if (hlen + cork->gso_size > cork->fragsize) { 1123 kfree_skb(skb); 1124 return -EINVAL; 1125 } 1126 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1127 kfree_skb(skb); 1128 return -EINVAL; 1129 } 1130 if (udp_sk(sk)->no_check6_tx) { 1131 kfree_skb(skb); 1132 return -EINVAL; 1133 } 1134 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1135 dst_xfrm(skb_dst(skb))) { 1136 kfree_skb(skb); 1137 return -EIO; 1138 } 1139 1140 skb_shinfo(skb)->gso_size = cork->gso_size; 1141 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1142 goto csum_partial; 1143 } 1144 1145 if (is_udplite) 1146 csum = udplite_csum(skb); 1147 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1148 skb->ip_summed = CHECKSUM_NONE; 1149 goto send; 1150 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1151 csum_partial: 1152 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1153 goto send; 1154 } else 1155 csum = udp_csum(skb); 1156 1157 /* add protocol-dependent pseudo-header */ 1158 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1159 len, fl6->flowi6_proto, csum); 1160 if (uh->check == 0) 1161 uh->check = CSUM_MANGLED_0; 1162 1163 send: 1164 err = ip6_send_skb(skb); 1165 if (err) { 1166 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1167 UDP6_INC_STATS(sock_net(sk), 1168 UDP_MIB_SNDBUFERRORS, is_udplite); 1169 err = 0; 1170 } 1171 } else { 1172 UDP6_INC_STATS(sock_net(sk), 1173 UDP_MIB_OUTDATAGRAMS, is_udplite); 1174 } 1175 return err; 1176 } 1177 1178 static int udp_v6_push_pending_frames(struct sock *sk) 1179 { 1180 struct sk_buff *skb; 1181 struct udp_sock *up = udp_sk(sk); 1182 struct flowi6 fl6; 1183 int err = 0; 1184 1185 if (up->pending == AF_INET) 1186 return udp_push_pending_frames(sk); 1187 1188 /* ip6_finish_skb will release the cork, so make a copy of 1189 * fl6 here. 1190 */ 1191 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1192 1193 skb = ip6_finish_skb(sk); 1194 if (!skb) 1195 goto out; 1196 1197 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1198 1199 out: 1200 up->len = 0; 1201 up->pending = 0; 1202 return err; 1203 } 1204 1205 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1206 { 1207 struct ipv6_txoptions opt_space; 1208 struct udp_sock *up = udp_sk(sk); 1209 struct inet_sock *inet = inet_sk(sk); 1210 struct ipv6_pinfo *np = inet6_sk(sk); 1211 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1212 struct in6_addr *daddr, *final_p, final; 1213 struct ipv6_txoptions *opt = NULL; 1214 struct ipv6_txoptions *opt_to_free = NULL; 1215 struct ip6_flowlabel *flowlabel = NULL; 1216 struct flowi6 fl6; 1217 struct dst_entry *dst; 1218 struct ipcm6_cookie ipc6; 1219 int addr_len = msg->msg_namelen; 1220 bool connected = false; 1221 int ulen = len; 1222 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1223 int err; 1224 int is_udplite = IS_UDPLITE(sk); 1225 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1226 1227 ipcm6_init(&ipc6); 1228 ipc6.gso_size = up->gso_size; 1229 ipc6.sockc.tsflags = sk->sk_tsflags; 1230 1231 /* destination address check */ 1232 if (sin6) { 1233 if (addr_len < offsetof(struct sockaddr, sa_data)) 1234 return -EINVAL; 1235 1236 switch (sin6->sin6_family) { 1237 case AF_INET6: 1238 if (addr_len < SIN6_LEN_RFC2133) 1239 return -EINVAL; 1240 daddr = &sin6->sin6_addr; 1241 if (ipv6_addr_any(daddr) && 1242 ipv6_addr_v4mapped(&np->saddr)) 1243 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1244 daddr); 1245 break; 1246 case AF_INET: 1247 goto do_udp_sendmsg; 1248 case AF_UNSPEC: 1249 msg->msg_name = sin6 = NULL; 1250 msg->msg_namelen = addr_len = 0; 1251 daddr = NULL; 1252 break; 1253 default: 1254 return -EINVAL; 1255 } 1256 } else if (!up->pending) { 1257 if (sk->sk_state != TCP_ESTABLISHED) 1258 return -EDESTADDRREQ; 1259 daddr = &sk->sk_v6_daddr; 1260 } else 1261 daddr = NULL; 1262 1263 if (daddr) { 1264 if (ipv6_addr_v4mapped(daddr)) { 1265 struct sockaddr_in sin; 1266 sin.sin_family = AF_INET; 1267 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1268 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1269 msg->msg_name = &sin; 1270 msg->msg_namelen = sizeof(sin); 1271 do_udp_sendmsg: 1272 if (__ipv6_only_sock(sk)) 1273 return -ENETUNREACH; 1274 return udp_sendmsg(sk, msg, len); 1275 } 1276 } 1277 1278 if (up->pending == AF_INET) 1279 return udp_sendmsg(sk, msg, len); 1280 1281 /* Rough check on arithmetic overflow, 1282 better check is made in ip6_append_data(). 1283 */ 1284 if (len > INT_MAX - sizeof(struct udphdr)) 1285 return -EMSGSIZE; 1286 1287 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1288 if (up->pending) { 1289 /* 1290 * There are pending frames. 1291 * The socket lock must be held while it's corked. 1292 */ 1293 lock_sock(sk); 1294 if (likely(up->pending)) { 1295 if (unlikely(up->pending != AF_INET6)) { 1296 release_sock(sk); 1297 return -EAFNOSUPPORT; 1298 } 1299 dst = NULL; 1300 goto do_append_data; 1301 } 1302 release_sock(sk); 1303 } 1304 ulen += sizeof(struct udphdr); 1305 1306 memset(&fl6, 0, sizeof(fl6)); 1307 1308 if (sin6) { 1309 if (sin6->sin6_port == 0) 1310 return -EINVAL; 1311 1312 fl6.fl6_dport = sin6->sin6_port; 1313 daddr = &sin6->sin6_addr; 1314 1315 if (np->sndflow) { 1316 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1317 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1318 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1319 if (!flowlabel) 1320 return -EINVAL; 1321 } 1322 } 1323 1324 /* 1325 * Otherwise it will be difficult to maintain 1326 * sk->sk_dst_cache. 1327 */ 1328 if (sk->sk_state == TCP_ESTABLISHED && 1329 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1330 daddr = &sk->sk_v6_daddr; 1331 1332 if (addr_len >= sizeof(struct sockaddr_in6) && 1333 sin6->sin6_scope_id && 1334 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1335 fl6.flowi6_oif = sin6->sin6_scope_id; 1336 } else { 1337 if (sk->sk_state != TCP_ESTABLISHED) 1338 return -EDESTADDRREQ; 1339 1340 fl6.fl6_dport = inet->inet_dport; 1341 daddr = &sk->sk_v6_daddr; 1342 fl6.flowlabel = np->flow_label; 1343 connected = true; 1344 } 1345 1346 if (!fl6.flowi6_oif) 1347 fl6.flowi6_oif = sk->sk_bound_dev_if; 1348 1349 if (!fl6.flowi6_oif) 1350 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1351 1352 fl6.flowi6_mark = sk->sk_mark; 1353 fl6.flowi6_uid = sk->sk_uid; 1354 1355 if (msg->msg_controllen) { 1356 opt = &opt_space; 1357 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1358 opt->tot_len = sizeof(*opt); 1359 ipc6.opt = opt; 1360 1361 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1362 if (err > 0) 1363 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1364 &ipc6); 1365 if (err < 0) { 1366 fl6_sock_release(flowlabel); 1367 return err; 1368 } 1369 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1370 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1371 if (!flowlabel) 1372 return -EINVAL; 1373 } 1374 if (!(opt->opt_nflen|opt->opt_flen)) 1375 opt = NULL; 1376 connected = false; 1377 } 1378 if (!opt) { 1379 opt = txopt_get(np); 1380 opt_to_free = opt; 1381 } 1382 if (flowlabel) 1383 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1384 opt = ipv6_fixup_options(&opt_space, opt); 1385 ipc6.opt = opt; 1386 1387 fl6.flowi6_proto = sk->sk_protocol; 1388 fl6.daddr = *daddr; 1389 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1390 fl6.saddr = np->saddr; 1391 fl6.fl6_sport = inet->inet_sport; 1392 1393 if (cgroup_bpf_enabled && !connected) { 1394 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1395 (struct sockaddr *)sin6, &fl6.saddr); 1396 if (err) 1397 goto out_no_dst; 1398 if (sin6) { 1399 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1400 /* BPF program rewrote IPv6-only by IPv4-mapped 1401 * IPv6. It's currently unsupported. 1402 */ 1403 err = -ENOTSUPP; 1404 goto out_no_dst; 1405 } 1406 if (sin6->sin6_port == 0) { 1407 /* BPF program set invalid port. Reject it. */ 1408 err = -EINVAL; 1409 goto out_no_dst; 1410 } 1411 fl6.fl6_dport = sin6->sin6_port; 1412 fl6.daddr = sin6->sin6_addr; 1413 } 1414 } 1415 1416 if (ipv6_addr_any(&fl6.daddr)) 1417 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1418 1419 final_p = fl6_update_dst(&fl6, opt, &final); 1420 if (final_p) 1421 connected = false; 1422 1423 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1424 fl6.flowi6_oif = np->mcast_oif; 1425 connected = false; 1426 } else if (!fl6.flowi6_oif) 1427 fl6.flowi6_oif = np->ucast_oif; 1428 1429 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1430 1431 if (ipc6.tclass < 0) 1432 ipc6.tclass = np->tclass; 1433 1434 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1435 1436 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1437 if (IS_ERR(dst)) { 1438 err = PTR_ERR(dst); 1439 dst = NULL; 1440 goto out; 1441 } 1442 1443 if (ipc6.hlimit < 0) 1444 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1445 1446 if (msg->msg_flags&MSG_CONFIRM) 1447 goto do_confirm; 1448 back_from_confirm: 1449 1450 /* Lockless fast path for the non-corking case */ 1451 if (!corkreq) { 1452 struct inet_cork_full cork; 1453 struct sk_buff *skb; 1454 1455 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1456 sizeof(struct udphdr), &ipc6, 1457 &fl6, (struct rt6_info *)dst, 1458 msg->msg_flags, &cork); 1459 err = PTR_ERR(skb); 1460 if (!IS_ERR_OR_NULL(skb)) 1461 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1462 goto out; 1463 } 1464 1465 lock_sock(sk); 1466 if (unlikely(up->pending)) { 1467 /* The socket is already corked while preparing it. */ 1468 /* ... which is an evident application bug. --ANK */ 1469 release_sock(sk); 1470 1471 net_dbg_ratelimited("udp cork app bug 2\n"); 1472 err = -EINVAL; 1473 goto out; 1474 } 1475 1476 up->pending = AF_INET6; 1477 1478 do_append_data: 1479 if (ipc6.dontfrag < 0) 1480 ipc6.dontfrag = np->dontfrag; 1481 up->len += ulen; 1482 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1483 &ipc6, &fl6, (struct rt6_info *)dst, 1484 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1485 if (err) 1486 udp_v6_flush_pending_frames(sk); 1487 else if (!corkreq) 1488 err = udp_v6_push_pending_frames(sk); 1489 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1490 up->pending = 0; 1491 1492 if (err > 0) 1493 err = np->recverr ? net_xmit_errno(err) : 0; 1494 release_sock(sk); 1495 1496 out: 1497 dst_release(dst); 1498 out_no_dst: 1499 fl6_sock_release(flowlabel); 1500 txopt_put(opt_to_free); 1501 if (!err) 1502 return len; 1503 /* 1504 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1505 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1506 * we don't have a good statistic (IpOutDiscards but it can be too many 1507 * things). We could add another new stat but at least for now that 1508 * seems like overkill. 1509 */ 1510 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1511 UDP6_INC_STATS(sock_net(sk), 1512 UDP_MIB_SNDBUFERRORS, is_udplite); 1513 } 1514 return err; 1515 1516 do_confirm: 1517 if (msg->msg_flags & MSG_PROBE) 1518 dst_confirm_neigh(dst, &fl6.daddr); 1519 if (!(msg->msg_flags&MSG_PROBE) || len) 1520 goto back_from_confirm; 1521 err = 0; 1522 goto out; 1523 } 1524 1525 void udpv6_destroy_sock(struct sock *sk) 1526 { 1527 struct udp_sock *up = udp_sk(sk); 1528 lock_sock(sk); 1529 udp_v6_flush_pending_frames(sk); 1530 release_sock(sk); 1531 1532 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1533 if (up->encap_type) { 1534 void (*encap_destroy)(struct sock *sk); 1535 encap_destroy = READ_ONCE(up->encap_destroy); 1536 if (encap_destroy) 1537 encap_destroy(sk); 1538 } 1539 if (up->encap_enabled) 1540 static_branch_dec(&udpv6_encap_needed_key); 1541 } 1542 1543 inet6_destroy_sock(sk); 1544 } 1545 1546 /* 1547 * Socket option code for UDP 1548 */ 1549 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1550 char __user *optval, unsigned int optlen) 1551 { 1552 if (level == SOL_UDP || level == SOL_UDPLITE) 1553 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1554 udp_v6_push_pending_frames); 1555 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1556 } 1557 1558 #ifdef CONFIG_COMPAT 1559 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1560 char __user *optval, unsigned int optlen) 1561 { 1562 if (level == SOL_UDP || level == SOL_UDPLITE) 1563 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1564 udp_v6_push_pending_frames); 1565 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1566 } 1567 #endif 1568 1569 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1570 char __user *optval, int __user *optlen) 1571 { 1572 if (level == SOL_UDP || level == SOL_UDPLITE) 1573 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1574 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1575 } 1576 1577 #ifdef CONFIG_COMPAT 1578 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1579 char __user *optval, int __user *optlen) 1580 { 1581 if (level == SOL_UDP || level == SOL_UDPLITE) 1582 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1583 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1584 } 1585 #endif 1586 1587 /* thinking of making this const? Don't. 1588 * early_demux can change based on sysctl. 1589 */ 1590 static struct inet6_protocol udpv6_protocol = { 1591 .early_demux = udp_v6_early_demux, 1592 .early_demux_handler = udp_v6_early_demux, 1593 .handler = udpv6_rcv, 1594 .err_handler = udpv6_err, 1595 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1596 }; 1597 1598 /* ------------------------------------------------------------------------ */ 1599 #ifdef CONFIG_PROC_FS 1600 int udp6_seq_show(struct seq_file *seq, void *v) 1601 { 1602 if (v == SEQ_START_TOKEN) { 1603 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1604 } else { 1605 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1606 struct inet_sock *inet = inet_sk(v); 1607 __u16 srcp = ntohs(inet->inet_sport); 1608 __u16 destp = ntohs(inet->inet_dport); 1609 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1610 udp_rqueue_get(v), bucket); 1611 } 1612 return 0; 1613 } 1614 1615 const struct seq_operations udp6_seq_ops = { 1616 .start = udp_seq_start, 1617 .next = udp_seq_next, 1618 .stop = udp_seq_stop, 1619 .show = udp6_seq_show, 1620 }; 1621 EXPORT_SYMBOL(udp6_seq_ops); 1622 1623 static struct udp_seq_afinfo udp6_seq_afinfo = { 1624 .family = AF_INET6, 1625 .udp_table = &udp_table, 1626 }; 1627 1628 int __net_init udp6_proc_init(struct net *net) 1629 { 1630 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1631 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1632 return -ENOMEM; 1633 return 0; 1634 } 1635 1636 void udp6_proc_exit(struct net *net) 1637 { 1638 remove_proc_entry("udp6", net->proc_net); 1639 } 1640 #endif /* CONFIG_PROC_FS */ 1641 1642 /* ------------------------------------------------------------------------ */ 1643 1644 struct proto udpv6_prot = { 1645 .name = "UDPv6", 1646 .owner = THIS_MODULE, 1647 .close = udp_lib_close, 1648 .pre_connect = udpv6_pre_connect, 1649 .connect = ip6_datagram_connect, 1650 .disconnect = udp_disconnect, 1651 .ioctl = udp_ioctl, 1652 .init = udp_init_sock, 1653 .destroy = udpv6_destroy_sock, 1654 .setsockopt = udpv6_setsockopt, 1655 .getsockopt = udpv6_getsockopt, 1656 .sendmsg = udpv6_sendmsg, 1657 .recvmsg = udpv6_recvmsg, 1658 .release_cb = ip6_datagram_release_cb, 1659 .hash = udp_lib_hash, 1660 .unhash = udp_lib_unhash, 1661 .rehash = udp_v6_rehash, 1662 .get_port = udp_v6_get_port, 1663 .memory_allocated = &udp_memory_allocated, 1664 .sysctl_mem = sysctl_udp_mem, 1665 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1666 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1667 .obj_size = sizeof(struct udp6_sock), 1668 .h.udp_table = &udp_table, 1669 #ifdef CONFIG_COMPAT 1670 .compat_setsockopt = compat_udpv6_setsockopt, 1671 .compat_getsockopt = compat_udpv6_getsockopt, 1672 #endif 1673 .diag_destroy = udp_abort, 1674 }; 1675 1676 static struct inet_protosw udpv6_protosw = { 1677 .type = SOCK_DGRAM, 1678 .protocol = IPPROTO_UDP, 1679 .prot = &udpv6_prot, 1680 .ops = &inet6_dgram_ops, 1681 .flags = INET_PROTOSW_PERMANENT, 1682 }; 1683 1684 int __init udpv6_init(void) 1685 { 1686 int ret; 1687 1688 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1689 if (ret) 1690 goto out; 1691 1692 ret = inet6_register_protosw(&udpv6_protosw); 1693 if (ret) 1694 goto out_udpv6_protocol; 1695 out: 1696 return ret; 1697 1698 out_udpv6_protocol: 1699 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1700 goto out; 1701 } 1702 1703 void udpv6_exit(void) 1704 { 1705 inet6_unregister_protosw(&udpv6_protosw); 1706 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1707 } 1708